Salient Feature Research Articles

Adopting multiple vision transformer layers for fine-grained image representation

Accurate discriminative regions proposal has an important effect for fine-grained image recognition. The vision transformer (ViT) brings about a striking effect in computer vision duo to its innate muti-head self-attention mechanism. However, the attention maps are gradually similar after certain layers and since ViT adds classification token for perform classification, it is unable to effectively select discriminative image patches for fine-grained image classification. To accurately detect discriminative regions, we propose a novel network AMTrans, which efficiently increases layers to learn diverse features and utilizes integrated raw attention maps to capture more salient feature. Specifically, we employ DeepViT as backbone to solve the attention collapse issue. Then, we fuse each head attention weight within each layer to produce attention weight map. After that, we alternatively use recurrent residual refinement blocks to promote salient feature detection and then utilize semantic grouping method to propose the discriminative feature region. A lot of experiments prove that AMTrans acquires the SOTA performance on three widely used fine-grained datasets under the same settings, involving Stanford-Cars, Stanford-Dogs and CUB-200-2011.

Multi-User Orbital Angular Momentum Based Terahertz Communications

Terahertz (THz) wireless communications are commonly regarded as one of the key technologies of 6G communication. Combined with THz, the newly exploited physical layer transmission dimension orbital angular momentum (OAM) that multiplexes a set of orthogonal modes on the same frequency channel, can unleash its potential in achieving high spectrum efficiency. Currently, most of the research on radio OAM communications focus on the point-to-point scenario in microwave and millimeter-wave (mmWave) bands. In this paper, we propose a uniform circular array (UCA)-based THz multi-user OAM (MU-OAM) communication system including downlink and uplink transmission schemes that simplifies the signal detection to the despiralization and amplitude detection (AD). A salient feature of the proposed MU-OAM communication scheme is lower computational complexity than the traditional MU-MIMO scheme without sacrifacing bit error rate (BER) and achievable sum rate performances, which is validated by mathematical analysis and numerical simulations.

On the dynamic mode-III crack in the elastic continuum consisting of sandy properties

The aim of the present investigation is to come up with a mathematical model for the dynamic mode III crack in an elastic continuum consisting of dry sandy properties. For the analysis of the propagating mode-III crack due to the propagating Love-type wave in an elastic continuum consisting of dry sandy properties, the results for the stress intensity factor (SIF) and crack opening displacement (COD) have been developed in closed form. The complex variable integral transform mathematical methods consisting of the two-sided Fourier integral transform, along with the Weiner-Hopf technique, serve as a salient feature of the adopted mathematical treatment for determining the SIF and COD in the closed form. Moreover, the expressions for the SIF and COD for the limiting static mode-III crack have also been obtained and deduced for special cases of elastic medium. The effects of existing parameters, viz. dry sandiness parameter, depth of crack, crack velocity on the SIF and COD . The effect of physical materials parameters has been shown numerically as well as graphically.

Systematic effects in the search for a muon electric dipole moment using the frozen-spin technique

At the Paul Scherrer Institute we are developing a high precision instrument to measure the electric dipole moment (EDM) of the muon. The salient feature of the experiment is the use of the frozen-spin technique to suppress the anomalous precession of the muon spin, allowing for a sensitivity that cannot be achieved with conventional g-2 muon storage rings. With this technique, the expected statistical sensitivity for the EDM after one year of data taking is 6 × 10-23 e · cm with a p = 125 MeV/c muon beam available at PSI.Reaching this goal necessitates a comprehensive analysis on spurious effects that mimic the EDM signal. This work discusses a quantitative approach to study systematic effects for the frozen-spin technique when searching for the muon EDM with a focus on the kinematics of decay positrons in the context of this experiment, which operates at muon momenta well below the ultrarelativistic limit.

Gender Differences in Facial Emotion Recognition Among Adolescents Depression with Non-Suicidal Self-Injury.

Despite the perception that healthy female are superior at emotional identification, it remains unclear whether gender-specific differences exist in adolescent depression and whether such specific differences in emotional recognition are associated with the most salient feature of adolescent depression---non-suicidal self-injury (NSSI). In this study, 1428 adolescents (1136 females and 292 males) with depression and NSSI were examined using the Facial Emotion Recognition Task, Patient Health Questionnaire-9 (PHQ-9), and Functional Assessment of Self-mutilation questionnaire (FASM). This study was grouped by gender. Data were analyzed using the descriptive statistics, independent sample t-test, chi-square test, non-parametric test (Mann-Whitney U-test), Spearman correlation and Multiple linear regression analysis. Depressed females reported a significantly greater frequency of self-injurious behaviour and more severe depressive symptoms than males. Face emotion recognition was also significantly more accurate in females and was positively correlated with levels of self-injury and depression, whereas no such correlations were found in males. Among depressed adolescents, face emotion recognition is better in females and is associated with self-injurious behaviour. This study found that the greater susceptibility to depression and NSSI among adolescent females may stem in part from superior recognition and sensitivity to the negative emotions of others.

Scale-Aware Tracking Method with Appearance Feature Filtering and Inter-Frame Continuity.

Visual object tracking is a fundamental task in computer vision that requires estimating the position and scale of a target object in a video sequence. However, scale variation is a difficult challenge that affects the performance and robustness of many trackers, especially those based on the discriminative correlation filter (DCF). Existing scale estimation methods based on multi-scale features are computationally expensive and degrade the real-time performance of the DCF-based tracker, especially in scenarios with restricted computing power. In this paper, we propose a practical and efficient solution that can handle scale changes without using multi-scale features and can be combined with any DCF-based tracker as a plug-in module. We use color name (CN) features and a salient feature to reduce the target appearance model's dimensionality. We then estimate the target scale based on a Gaussian distribution model and introduce global and local scale consistency assumptions to restore the target's scale. We fuse the tracking results with the DCF-based tracker to obtain the new position and scale of the target. We evaluate our method on the benchmark dataset Temple Color 128 and compare it with some popular trackers. Our method achieves competitive accuracy and robustness while significantly reducing the computational cost.

Hippocampal subfield vulnerability to α-synuclein pathology precedes neurodegeneration and cognitive dysfunction

Cognitive dysfunction is a salient feature of Parkinson’s disease (PD) and Dementia with Lewy bodies (DLB). The onset of dementia reflects the spread of Lewy pathology throughout forebrain structures. The mere presence of Lewy pathology, however, provides limited indication of cognitive status. Thus, it remains unclear whether Lewy pathology is the de facto substrate driving cognitive dysfunction in PD and DLB. Through application of α-synuclein fibrils in vivo, we sought to examine the influence of pathologic inclusions on cognition. Following stereotactic injection of α-synuclein fibrils within the mouse forebrain, we measured the burden of α-synuclein pathology at 1-, 3-, and 6-months post-injection within subregions of the hippocampus and cortex. Under this paradigm, the hippocampal CA2/3 subfield was especially susceptible to α-synuclein pathology. Strikingly, we observed a drastic reduction of pathology in the CA2/3 subfield across time-points, consistent with the consolidation of α-synuclein pathology into dense somatic inclusions followed by neurodegeneration. Silver-positive degenerating neurites were observed prior to neuronal loss, suggesting that this might be an early feature of fibril-induced neurotoxicity and a precursor to neurodegeneration. Critically, mice injected with α-synuclein fibrils developed progressive deficits in spatial learning and memory. These findings support that the formation of α-synuclein inclusions in the mouse forebrain precipitate neurodegenerative changes that recapitulate features of Lewy-related cognitive dysfunction.

Playing with your ears: Audio-motor skill learning is sensitive to the lateral relationship between trained hand and ear

A salient feature of motor and sensory circuits in the brain is their contralateral hemispheric bias-a feature that might play a role in integration and learning of sensorimotor skills. In the current behavioral study, we examined whether the lateral configuration between sound-producing hand and feedback-receiving ear affects performance and learning of an audio-motor skill. Right-handed participants (n=117) trained to play a piano sequence using their right or left hand while auditory feedback was presented monaurally, either to the right or left ear. Participants receiving auditory feedback to the contralateral ear during training performed better than participants receiving ipsilateral feedback (with respect to the training hand). Furthermore, in the Left-Hand training groups, the contralateral training advantage persisted in a generalization task. Our results demonstrate that audio-motor learning is sensitive to the lateral configuration between motor and sensory circuits and suggest that integration of neural activity across hemispheres facilitates such learning.

Voxelized atomic structure framework for materials design and discovery

We present a computational framework for developing physics-based, high-fidelity structure–property relationships with atomic systems. In this framework, atomic structure is quantified by directionally resolved two-point spatial correlations of the charge density field, projected to a salient low-dimensional feature space via principal component analysis (PCA), and correlated to physical properties by Gaussian process regression (GPR). The charge density field provides a complete, purely physics-based definition of the atomic structure that is independent of chemical species information and does not require additional feature engineering or idealizations beyond those of first-principles computations. The two-point spatial correlations capture the salient spatial features underlying the atomic structure that dictate the physics underlying the material response. Since the feature engineering approach explored in this work is universally applicable to all atomic structures independent of the chemical species present in the structure, it offers new avenues for efficiently exploring the space of atomic structures for desired property combinations. A further contribution of this work comes from utilizing the uncertainty quantification inherently provided by GPR to deploy a Bayesian experiment design strategy to minimize the number of computationally expensive physics simulations required to achieve the desired accuracy. In this work, we demonstrate the proposed framework to elucidate the relationship between the chemical composition and bulk modulus in AlNbTiZr high entropy alloys. It is shown that a highly accurate structure–property relationship with less than 2% average error can be established using a small training dataset of less than 30 samples.

Dielectric diatomic metasurface-assisted versatile bifunctional polarization conversions and incidence-polarization-secured meta-image.

Dielectric metasurface empowering efficient light polarization control at the nanoscale, has recently garnered tremendous research interests in the field of high-resolution image encryption and display, particularly at low-loss wavelengths in the visible band. Nevertheless, due to the single fixed polarization conversion function, the image (either positive or negative image) can always be decrypted in a host-uncontrollable manner as long as the user applies an analyzer to select the polarization component of the output light. Here, we resort to half-waveplate- and quarter-waveplate-like silicon nanopillars to form a metamolecule of a dielectric diatomic metasurface, which can yield versatile linearly polarized (LP) and circularly polarized (CP) light upon orthogonally linear-polarized incidences, providing new degrees of freedom for image display and encryption. We show both theoretically and numerically that versatile different paired LP and CP combinations could be achieved by simply adjusting the orientation angles of the two nanopillars. The bifunctional polarization conversion functions make possible that a meta-image can only be seen when incident light is linearly polarized at a specific polarization angle, whereas no image can be discerned for the orthogonal polarization incidence case, indicating the realization of incidence-polarization secured meta-image. This salient feature holds for all individual metamolecules, reaching a remarkable image resolution of 52,916 dots per inch. By fully exploiting all polarization conversions of four designed metamolecules, three-level incidence polarization-secured meta-image can also be expected.

Small samples-oriented intrinsically explainable machine learning using Variational Bayesian Logistic Regression: An intensive care unit readmission prediction case for liver transplantation patients

Recently, machine learning has revolutionized medical diagnosis and prediction, rivaling human experts in accuracy. However, the limited interpretability poses challenges in meeting stringent medical decision standards. To facilitate interpretable predictive inference, this article proposes a small samples-oriented intrinsically explainable machine learning model, an Optimal Variational Bayesian Logistic Regression (OVBLR) model with a salient feature estimation strategy (OVBLR-SFE). OVBLR-SFE sequentially learns from limited labeled samples, mining disease-related risk factors to provide intrinsic interpretability to its predictive inference automatically. The model achieves this by reasonably approximating the posterior probability through the proposed OVBLR model to estimate the feature weight and influential effect on model outputs to facilitate salient feature selection. The effectiveness and superiority of OVBLR-SFE are demonstrated through extensive experiments on four benchmark medical datasets from the UCI repository, including the Bone Marrow Transplant: Children Data Set, Breast Cancer Wisconsin (Original) Data Set, SPECT Heart Data Set, and Heart Disease Data Set, as well as a real-world case of intensive care unit readmission prediction. Experimental results reveal that the OVBLR-SFE can achieve high accuracies of 92.02%, 97.21%, 86.71%, and 84.47%, respectively, with an average classification accuracy of 90.10% stably on the four UCI datasets. Especially, OVBLR-SFE can accurately infer significant features that align with that obtained by the post-hoc interpretation approaches: SHAP and AcME, demonstrating remarkable interpretability. In the real application, the proposed method achieved a readmission prediction accuracy of 88.11% in predicting the readmission probabilities of liver transplantation patients. Notably, the inferred salient features, such as total Intensive Care Unit length of stay (ICUTime), N-terminal pro-brain natriuretic peptide(NTproBNP), Operating time of surgery(OperationTime_h), and intraoperative red blood cell transfusion (RBCT), emerges as crucial in readmission prediction, aligning with the assessments of clinical experts. The source code of this study can be accessed at https://github.com/xqw42/OVBLR-SFE.

S/kinship: The relational ontology of tattoos in contemporary Australian discourse and practice

AbstractIn examining affective motivations and meanings associated with kin‐based tattooing, this article proposes a practice of ‘s/kinship’: the tangible inscription of relational personhood on the body. While research on the practice of Western tattooing has long been drawn into discourses of deviance and individual identity, little attention has been paid to the tattooing of social relations. Drawing on surveys and interview data, this article highlights the significance of kinship and other forms of relationality in tattooing, bringing attention to how these are manifested through the skin as expressions of social proximity, permanence, and porosity. In doing so an analytic approach to s/kinship is developed to foreground the practice of tattooing as an expression of relational personhood, one that highlights the permeability of personhood as a salient ontological feature in the social life of tattoos and the tattooed.

Neuromodulatory functions (reward and arousal) induce separate effects on visual perceptual learning (VPL) of a salient but goal-irrelevant visual feature

Neuromodulatory functions (reward and arousal) induce separate effects on visual perceptual learning (VPL) of a salient but goal-irrelevant visual feature

Comparative performance of nonlinear energy harvesters through strongly coupled fluid-structure-electrical interactive models

Micro power generators (MPGs) can harvest electrical power from ambient energy resources in natural environments to operate small-scale electronic devices. Such systems exploit structural vibrations due to cross-flow instabilities through fluid–structure interactions (FSI). Due to attractive features of piezoelectric transducers, we utilize them for vibration-to-electricity conversion in this work. We develop strongly coupled FSI-based computational models to handle large-amplitude structural motions in which unsteady forces exerted by fluid flows on structures, their dynamical responses, and harvested power are directly computed. Our mathematical model for incompressible flows over moving bodies relies on arbitrary Lagrangian–Eulerian (ALE) formulation. Our in-house FSI solver primarily works using radial basis interpolation technique for grid morphing. Using broad ranges of fluidic, structural, and electrical parameters, we investigate the power generation performance of two types of MPGs, working through vortex-induced vibrations of a circular cylinder and a fluttering NACA-0012 airfoil, under different excitation conditions. Optimum values of load resistance are found to maximize the harvested power. Furthermore, our findings support the linear analysis, which provides insights into the possible instabilities in the electro-mechanical system. Another salient feature of this research work is the inclusion of cubic nonlinear stiffness in the energy harvesting models.

Predictor-Based Data-Driven Model-Free Adaptive Predictive Control of Power Converters Using Machine Learning

In this article, a novel robust data-driven model-free predictive control framework based on the I/O data of the controlled plants, which is performed by incorporating the neural predictor-based model-free adaptive control and finite control-set model predictive control, is first proposed. The salient feature of the suggested framework is that the uncertainties, such as unmodeled dynamics and external disturbances, can be explicitly addressed in controlled systems. From a practical standpoint, however, the potential of this proposal is limited by a significantly increased online computational complexity, which makes it difficult to implement. To circumvent this limitation, a supervised imitation learning technique using data labeled is developed to imitate the known suggested controller, which the majority of the online computational burden can be transformed into offline computing by utilizing a trained artificial neural network subject to robustness characteristics. In particular, this development motivates a much simpler robust predictive control solution, which is convenient to implement in applications. Thus, by this proposal, the online implementation of much more complex predictive control strategies is made possible, and it explores a new possibility for future development of the complex control methodology. Finally, extensive simulative and experimental investigations for modular multilevel converter validate the interest and viability of the proposed design methodology.

Tracking of Blood Vessels Motion from 4D-flow MRI Data.

This paper presents a novel approach to track objects from 4D-flow MRI data. A salient feature of the proposed method is that it fully exploits the geometrical and dynamical nature of the information provided by this imaging modality. The underlying idea consists in formulating the tracking problem as a data assimilation problem, in which both position and velocity observations are extracted from the 4D-flow MRI data series. Optimal state estimation is then performed in a sequential fashion via Kalman filtering. The capabilities of the method are extensively assessed in a numerical study involving synthetic and clinical data.

Nonparametric clustering of RNA‐sequencing data

AbstractIdentification of clusters of co‐expressed genes in transcriptomic data is a difficult task. Most algorithms used for this purpose can be classified into two broad categories: distance‐based or model‐based approaches. Distance‐based approaches typically utilize a distance function between pairs of data objects and group similar objects together into clusters. Model‐based approaches are based on using the mixture‐modeling framework. Compared to distance‐based approaches, model‐based approaches offer better interpretability because each cluster can be explicitly characterized in terms of the proposed model. However, these models present a particular difficulty in identifying a correct multivariate distribution that a mixture can be based upon. In this manuscript, we review some of the approaches used to select a distribution for the needed mixture model first. Then, we propose avoiding this problem altogether by using a nonparametric MSL (maximum smoothed likelihood) algorithm. This algorithm was proposed earlier in statistical literature but has not been, to the best of our knowledge, applied to transcriptomics data. The salient feature of this approach is that it avoids explicit specification of distributions of individual biological samples altogether, thus making the task of a practitioner easier. We performed both a simulation study and an application of the proposed algorithm to two different real datasets. When used on a real dataset, the algorithm produces a large number of biologically meaningful clusters and performs at least as well as several other mixture‐based algorithms commonly used for RNA‐seq data clustering. Our results also show that this algorithm is capable of uncovering clustering solutions that may go unnoticed by several other model‐based clustering algorithms. Our code is publicly available on Github at https://github.com/Matematikoi/non_parametric_clustering

Coupled meshfree (SPH) and grid based (FDM) procedures for modeling fluid flow through deformable porous media

A coupled numerical framework based on smoothed particle hydrodynamics (SPH) and finite difference method (FDM) has been developed to analyze the coupled process of fluid flow in a deformable poro-elastic medium. The salient feature of the developed coupled model is that the porous media has been simulated by two separate layers to describe solid and fluid phase with their respective governing equations and solved simultaneously with mutual transfer of variables from one phase to another. SPH being meshfree method has advantage over grid based methods to simulate large deformations and fracture propagation and hence used for simulating the solid phase while the pore pressure variation in the fluid phase has been analyzed by FDM. Detailed mathematical framework and procedure explaining the mutual exchange of information between SPH and FDM has been discussed in this paper. The change of porosity due to solid deformation and its impact on variation of fluid pressure has also been included in the proposed model. The developed coupled model has been finally validated with three standard benchmark problems by comparing numerical results of the proposed model with those of analytical solutions, to establish its accuracy and versatility in modeling fluid flow through deformable porous rock media.

Small arteriole sign: an imaging feature for staging T4a colon cancer.

By analyzing the distribution of existing and newly proposed staging imaging features in pT1-3 and pT4a tumors, we searched for a salient feature and validated its diagnostic performance. Preoperative multiphase contrast-enhanced CT images of the training cohort were retrospectively collected at three centers from January 2016 to December 2017. We used the chi-square test to analyze the distribution of several stage-related imaging features in pT1-3 and pT4a tumors, including small arteriole sign (SAS), outer edge of the intestine, tumor invasion range, and peritumoral adipose tissue. Preoperative multiphase contrast-enhanced CT images of the validation cohort were retrospectively collected at Beijing Cancer Hospital from January 2018 to December 2018. The diagnostic performance of the selected imaging feature, including accuracy, sensitivity, and specificity, was validated and compared with the conventional clinical tumor stage (cT) by the McNemar test. In the training cohort, a total of 268 patients were enrolled, and only SAS was significantly different between pT1-3 and pT4a tumors. The accuracy, sensitivity, and specificity of the SAS and conventional cT in differentiating T1-3 and T4a tumors were 94.4%, 81.6%, and 97.3% and 53.7%, 32.7%, and 58.4%, respectively (all p < 0.001). In the validation cohort, a total of 135 patients were collected. The accuracy, sensitivity, and specificity of the SAS and the conventional cT were 93.3%, 76.2%, and 96.5% and 62.2%, 38.1%, and 66.7%, respectively (p < 0.001, p = 0.021, p < 0.001). Small arteriole sign positivity, an indirect imaging feature of serosa invasion, may improve the accuracy of identifying T4a colon cancer. Small arteriole sign helps to distinguish T1-3 and T4a colon cancer and further improves the accuracy of preoperative CT staging of colon cancer. • The accuracy of preoperative CT staging of colon cancer is not ideal, especially for T4a tumors. • Small arteriole sign (SAS) is a newly defined imaging feature that shows the appearance of tumor-supplying arterioles at the site where they penetrate the intestine wall. • SAS is an indirect imaging marker of tumor invasion into the serosa with a great value in distinguishing between T1-3 and T4a colon cancer.

Video Super-Resolution Network with Gated High-Low Resolution Frames

In scenes with large inter-frame motion variations, distant targets, and blurred targets, the lack of inter-frame alignment can greatly affect the effectiveness of subsequent video super-resolution reconstruction. How to perform inter-frame alignment in such scenes is the key to super-resolution reconstruction. In this paper, a new motion compensation method is proposed to design an alignment network based on gated high-low resolution frames. The core idea is to introduce a gating mechanism while using the information of high-low resolution neighboring frames to perform motion compensation adaptively. Meanwhile, within this alignment framework, we further introduce a pre-initial hidden state network and a local scale hierarchical salient feature fusion network. The pre-initial hidden state network is mainly used to reduce the impact of unbalanced quality effects between frames that occur in one-way cyclical networks; the local scale hierarchical salient feature fusion network is used to fuse the features of aligned video frames to extract contextual information and locally salient features to improve the reconstruction quality of the video. Compared with existing video super-resolution methods, this method achieves good performance and clearer edge and texture details.